Two circles, one center, and the thin space between them. Fill that space with Möbius chips — plus one, minus one, and hole — and watch the gap breathe.
Build one circle out of the primes and it comes out perfect — every chip laid evenly, a smooth ring of infinite fuel. Now deal the primes into two circles instead of one, and a gap opens between them. That gap is a ledger. Whatever you write into it, its running total decides how far the outer ring drifts from the inner. So the only question left is: what do you write in the gap?
The forward woodchipper only ever throws one kind of chip — positive, additive, a pile that grows and never cancels. That is why it cannot be run in reverse: nothing in it can take anything away. The reverse needs three chips, not one. It needs +1, it needs −1, and it needs the chip a grinding machine can never make: 0 — the empty seat, the number that holds a place and casts no vote. Those three tokens already have a name. They are the Möbius function, μ(n): +1 for a squarefree number with an even count of prime factors, −1 for an odd count, and 0 for anything divisible by a square — anything not clean.
Lay one μ-chip in the gap for every integer, going around. The running sum of those chips is the width of the gap at each step — and that running sum is Mertens's function, M(n) = μ(1)+μ(2)+…+μ(n). Below is the machine doing exactly that. The pale circle is the ideal — perfectly concentric, the circle the primes would build if they were tame. The wandering line is what they actually build. Where it swings out past the ideal, M is positive and the gap glows warm; where it dips inside, M is negative and the gap goes cool; where the two curves kiss, M has returned to zero and the gap has pinched shut.
Each spoke is one integer, dealt clockwise from the top. n runs 1 → 3000 around the ring.
Everything the ring draws is real. μ(n) is a real function; M(n) is its real running total; the pinch-points really are the integers where M returns to zero. The wander really does stay startlingly small for a machine fed three thousand chips — that smallness is the whole mystery.
For a long time people guessed the gap could never widen past √n — the Mertens conjecture, |M(n)| < √n. It is false. Odlyzko and te Riele disproved it in 1985; somewhere out past the range any machine will ever draw, the gap does breach that wall — though the smallest n where it happens still isn't known. What survives is softer and deeper: the Riemann Hypothesis is equivalent to the gap growing no faster than M(x) = O(x^{1/2+ε}). So the width of this gap, taken all the way out, is not a curiosity. It is one face of the last great open question about the primes.
Which is the honest end of the woodchipper run backward. You can't un-grind the pile. But you can build the machine that would have to run in reverse — the one with the minus and the hole — and let it draw you the exact shape of its own impossibility, breathing, on a ring.
the-wander · onojk123.com · μ(n), M(n), and the space between two circles
math is honest where it is math and a picture where it is only a picture